Optimal Control to Limit the Propagation Effect of a Virus Outbreak on
a Network
Paolo Di Giamberardino and Daniela Iacoviello
Dept. Computer, Control and Management Engineering Antonio Ruberti, Sapienza University of Rome,
via Ariosto 25, 00185 Rome, Italy
Keywords:
Epidemic Modeling, Computer Virus, System Analysis, Optimal Control.
Abstract:
The aim of this paper is to propose an optimal control strategy to face the propagation effects of a virus
outbreak on a network; a recently proposed model is integrated and analysed. Depending on the specific
model caracteristics, the epidemic spread could be more or less dangerous leading to a virus free or to a virus
equilibrium. Two possible controls are introduced: a test on the computers connected in a network and the
antivirus. In a condition of limited resources the best allocation strategy should allow to reduce the spread of
the virus as soon as possible.
1 INTRODUCTION
The computer virus spread represents an important is-
sue since all the connected devices are susceptible of
being infected. A computer virus is a code that in gen-
eral can modify normal operations, as well as dam-
age files, and attack other computers. It can be trans-
mitted by downloading files from internet, by using
external devices, or by e-mails. The implications of
this threat involve the field of Cyber Security, that has
been recently defined as a game between defender and
attacker, (Karunanithi et al., 2018). Since the com-
puters and the internet connection are becoming more
and more widespread, a computer virus is able to dis-
rupt the productivity and cause billions of damages,
(Zhu and Yang, 2012); therefore, a large amount of
resources has been dedicated to blocking the spread
of viruses.
There are many analogies with diseases epidemic
spread and also the nomenclature is similar. The ba-
sic model is the SIR one, where S stands for suscep-
tible, indicating the subjects, in this case the comput-
ers, free from virus but that can be infected, I stands
for the set of all the computers infected and infectious
and R represents the set of all recovered computers.
The computer virus may have a latent period dur-
ing which the device, while being infected, is still not
able to infect other computers, (Peng et al., 2013);
the delay with which a virus breaks out is an im-
portant parameter to be taken into account when im-
plementing a control strategy. From the early 1980s
the problem of computer virus detection and removal
has been faced, taking into account the characteris-
tics of this kind of infection: the latency, the para-
sitism, the hiding and the infectiousness (Hu et al.,
2015). The modeling of the computers virus dynam-
ics can vary depending on the specific scenario con-
sidered that suggests a more or less complex par-
tition of the population. In the recent paper (Fa-
tima et al., 2018) a susceptible-latent-breakingout-
quarantine-susceptible computer virus dynamics is
proposed and implemented, showing the positive ef-
fects of the quarantine. The scenario considered in
this paper is similar to the one in (Xu and Ren, 2016);
the population of computers connected in a network
is partitioned into four classes. Besides the classes
of susceptible S and recovered R computers, that are
the devices not infected that can become infected and
those that can not get the infection respectively, there
are two classes of infected computers. The first one,
E, is the most dangerous, since it is assumed that the
virus has not yet manifested itself but the computers
in this class can infect the devices they get in touch
with; the second class, I, contains the computers in-
fected and in which the virus has broken itself out.
This is a common condition that could include in-
fected e-mails or viruses that break out with a delay.
It appears important to become aware of the presence
of a virus as soon as possible, and successively to ap-
ply the suitable antivirus. This is the rationale for the
choices of the two strategies proposed: u
1
represents
the test to be performed on the computers which are
Di Giamberardino, P. and Iacoviello, D.
Optimal Control to Limit the Propagation Effect of a Virus Outbreak on a Network.
DOI: 10.5220/0008052804550462
In Proceedings of the 16th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2019), pages 455-462
ISBN: 978-989-758-380-3
Copyright
c
2019 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
455
supposed free of virus, i.e. S and E, while u
2
repre-
sents the antivirus to be applied on the known infected
computers I, once the infection has broken out.
In this paper a variation of the model proposed in
(Xu and Ren, 2016) is introduced and analysed, by
considering the reproduction number and the possible
existence of virus equilibrium. Optimal control strate-
gies are proposed to face a possible epidemic spread.
The paper is organized as follows; Section 2 is divided
into three parts; in the first two a new model describ-
ing the computer virus spread is analysed considering
the reproduction number, the equilibrium points and
their stability. In the third, the optimal control strategy
is proposed to allocate efficiently the limited available
resources. In Section 3 numerical results show the
plausibility of the model and the effectiveness of the
introduced actions. Conclusions and future develop-
ments are presented in Section 4.
2 MATERIALS AND METHODS
In this Section a new model of virus spread in a com-
puter network is described and analysed. The inspi-
ration is the model proposed in the paper in (Xu and
Ren, 2016); the total set of computers connected the
internet is partitioned into four classes: S, the com-
partment of susceptible computers: they are the un-
infected computers connected to the network; E, the
compartment of exposed computers: they are the in-
fected computers where all viruses are latent; I, the
compartment of infected computers: they are the in-
fected computers where all the viruses are currently
breaking out; R, the compartment of recovered com-
puters: they are the recovered computers that have run
the antivirus software.
It is assumed that new computers are connected to
the network with rate b; a susceptible computer in S,
when having a connection with a computer in the ex-
posed class E, can get the virus with rate β. As an ef-
fect, it can become infected, moving to I, with a prob-
ability p, or latent, moving to E, with a probability
1 p. A latent computer virus breaks out with rate γ,
and the computers are removed from the net with rate
µ; moreover, it is possible that the virus is temporar-
ily suppressed with probability ε > 0. The proposed
model is a variation of the one in (Xu and Ren, 2016),
having modeled the infection interaction between sus-
ceptible and exposed computers by the term βSE and
not βSI; another difference is the impossibility for an
infected device to become recovered without an ex-
ternal control.
The control actions introduced in this paper to
limit the spread of a virus consist in a prevention cam-
paign u
1
, such as to induce people to test the condi-
tion and the vulnerability of their computer, and in
the antivirus action u
2
; they are weighted by σ
1
and
σ
2
representing the unit cost efficiency. The virus test
u
1
is applied on all the susceptible computers in S and
its result influences the evolution of number of the ex-
posed computers. The proposed model completed by
the control actions is:
˙
S = b βSE µS (1)
˙
E = (1 p)βSE + εI (γ+ µ)E σ
1
(
E
S + E
)u
1
(2)
˙
I = pβSE + γE (ε + µ)I + σ
1
(
E
S + E
)u
1
σ
2
Iu
2
(3)
˙
R = µR + σ
2
Iu
2
(4)
The controls u
i
are assumed bounded:
0 u
i
(t) U
M
i
, i = 1, 2 (5)
being U
M
i
the corresponding possible maximum value
of u
i
. Obviously, all the state variables S, E, I and
R are non negative, representing the number of the
computers in the four identified different conditions;
then, the same must hold for the initial conditions
S(0) = S
0
0 E(0) = E
0
0
I(0) = I
0
0 R(0) = R
0
0 (6)
2.1 The Model Analysis
The proposed model is now analysed to understand its
dynamical behavior. Two related aspects are now con-
sidered; the determination of the equilibrium points
and the reproduction number. To determine the equi-
librium points, the equation (4) of the model is not
informative and therefore only the first three are con-
sidered from now on. For sake of notation, the model
(1)–(3) is represented as follows:
˙
X = f (X,U) (7)
with state X =
X
1
X
2
X
3
T
=
S E I
T
and control input U =
u
1
u
2
T
In (7), f =
f
1
f
2
f
3
T
, where the f
i
are the r.h.s. functions
of equations (1)–(3). The equilibrium points are de-
termined by assuming null the control inputs u
i
and
considering the equation f (X) = 0. One solution, that
can be referred as the disease free equilibrium, is
X
e1
=
b
µ
0 0
T
(8)
To determine the other equilibrium points, if they
exist, it is useful to introduce the auxiliary variable
Z = βE + µ; from f
1
= 0
S =
b
Z
, (9)
ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics
456
can be obtained, whereas from f
3
= 0 one has
I =
E(pβb + γZ)
Z(ε + µ)
(10)
Therefore, the existence of S and I as equilibrium
state components depends on the existence of E. By
substituting expressions (9) and (10) in f
2
= 0, one
obtains E = 0, the disease free equilibrium, and
Z =
bβ(ε + µ )
µ(ε + µ + γ)
(11)
By recalling the definition of Z, one obtains:
E =
b(ε + µ )
µ(ε + µ + γ)
µ
β
=
bA
µB
µ
β
(12)
once
A = ε + µ(1 p) B = ε + µ + γ (13)
are defined. If E > 0, that is
bA
µB
>
µ
β
(14)
then a second equilibrium point is obtained:
X
e2
=
µB
βA
bA
µB
µ
β
(bAβµ
2
B)(pµB+Aγ)
µBβA(µ+ε)
(15)
It is useful to connect the existence of the second equi-
librium point with the rate β at which computers be-
come infected. From (14) it can be deduced that if
β >
µ
2
B
bA
= β
b
(16)
then system (7) has two equilibrium points, X
e1
and
X
e2
; if β < β
b
, the solution is not feasible since the
components assume negative values; for β = β
b
the
two solutions X
e1
and X
e2
coincide. The value β
b
is
the bifurcation one.
Now local stability of the equilibrium points is in-
vestigated computing the Jacobian matrix of the sys-
tem
J =
βE µ βS 0
(1 p)βE (1 p)βS (γ + µ) ε
pβE pβS + γ (ε + µ)
(17)
and evaluating it in X
e1
and, if it exists, in X
e2
. As far
as the point X
e1
in (8) is concerned, the evaluation of
(17) gives
J(X
e1
) =
µ β
b
µ
0
0 (1 p)β
b
µ
(γ + µ) ε
0 pβ
b
µ
+ γ (ε + µ)
(18)
One eigenvalue of (18) is clearly λ
1
= µ; the
other two are the roots of the polynomial equation
λ
2
+
(ε + γ + 2µ) (1 p)β
b
µ
λ+
µ(ε + γ + µ) β
b
µ
(ε + µ(1 p)) = 0 (19)
The two solutions, written as function of all the pa-
rameters appearing in the equation, do not allow an
easy analysis. However, once the bifurcation value
(16) is computed, it can be interesting to study the
sign of the roots of (19) in a neighbourhood of the bi-
furcation condition. The two coefficients can be eval-
uated setting β = β
b
± δ, with 0 < δ 1. By using
the expression (16) for β
b
, as well as the notations in
(13), the polynomial equation (19) becomes
λ
2
+
ε
B
A
+ µ (1 p)
b
µ
δ
λ+
(ε + µ(1 p))
b
µ
δ = 0 (20)
It can be noted that if β = β
b
δ is considered, the
two coefficients of (20) are
ε
B
A
+ µ + (1 p)
b
µ
δ > 0 (21)
(ε + µ(1 p))
b
µ
δ > 0 (22)
Then, for the Descartes’ rule of signs, the two solu-
tions have negative real part and the equilibrium point
is locally asymptotically stable. Once β = β
b
+ δ is
assumed, the coefficient (22) becomes
(ε + µ(1 p))
b
µ
δ < 0 (23)
while the coefficient of λ changes its sign as δ varies.
However, no matter what the sign of the coefficient of
λ is: the polynomial equation has one positive and one
negative real solutions; consequently, the equilibrium
point X
e1
results unstable. This behaviour follows
the classical case of generic epidemic spread: there
is a bifurcation condition which separates the case of
only one asymptotically stable equilibrium point, cor-
responding to the disease free condition, and the case
of two equilibria, one again the disease free condition
which becomes unstable, and the second equilibrium
point, corresponding to the so called endemic condi-
tion, locally asymptotically stable. Also in this case it
is possible to verify that the equilibrium point X
e2
is
locally asymptotically stable, when it exists. Since the
conditions obtained are function of all the parameters
involved, their analytical expressions are quite com-
plicated and not useful to bring to clear conclusions.
In the numerical simulations, after the definition of
the parameters values, the cases β < β
b
and β > β
b
are verified and illustrated.
Optimal Control to Limit the Propagation Effect of a Virus Outbreak on a Network
457
2.2 The Reproduction Number
In epidemic spread the reproduction number R , as
recalled in (Driessche, 2002) and (Driessche, 2017),
is a useful parameter to describe the ability of an in-
fectious disease to invade a population. In this case
the population is the set of all computers connected in
the network and R is now evaluated by the next gen-
eration matrix. More precisely, from the dynamical
model (1)– (3), considering only the compartments
containing the infected subjects, the dynamical evolu-
tions (2) and (3) can be split into two parts, collected
in the two vectors:
K =
(1 p)βSE
pβSE
Y =
(µ + γ)E εI
(ε + µ)I γE
The matrix K accounts for the rate of appearance of
new infections in the compartments E and I, whereas
Y describes the rate of other transitions between them.
Now, the jacobian of K and Y (with respect to E and
I) evaluated at the free disease equilibrium X
e1
yields
respectively:
F =
(1p)βb
µ
0
pβb
µ
0
!
V =
µ + γ ε
γ ε + µ
The next generation matrix is FV
1
, that is:
FV
1
=
βb
(µ + ε)(1 p) ε(1 p)
p(ε + µ) pε
µ[(µ + γ)(ε + µ) εγ]
The reproduction number is defined as the spectral
radius of the matrix FV
1
; in this case it is quite easy
to compute it, getting
R =
(ε + µ )βb
µ[(µ + γ)(ε + µ) εγ]
=
Aβb
µ
2
B
(24)
Following the classical definition of reproduction
number, when R < 1 the epidemic spread is charac-
terised by a low contagious effect which tends to keep
limited the number of infected units, asymptotically
going to the disease free equilibrium; in the cases of
R > 1 the number of infected machines grows, go-
ing asymptotically to an endemic condition where the
more R is great the more are the infected machines.
In the present case, comparing expression (24) with
(16), it is possible to find the relationship
R =
β
β
b
(25)
between the reproduction number and the bifurcation
condition, so that R > 1 coincides with the condition
of existence of the second equilibrium point X
e2
, (14).
2.3 The Optimal Control Problem
Solution
In the proposed model (1)–(4) two limited controls
are introduced, u
1
and u
2
, aiming at a fast virus de-
tection and a suitable antivirus action; with the first
control the computers in the infected latent condition
(E) are detected, whereas with the second one the an-
tivirus allows the recovery of the infected computers
in I. The available resources are generally bounded,
from economic, logistic and practical point of view;
this implies the need of suitable allocation strategy.
In the similar contest of epidemic diseases, the natu-
ral framework to face efficiently the virus spread with
limited resources is the optimal control theory. A cost
index is introduced; the idea is to choose the control
in such a way that the total number of infected com-
puters (i.e. computers in E and in the I classes) is
minimized, taking into account the resources limita-
tions (5). For sake of notation, they are rewritten as
q
1
= u
1
0 q
2
= u
1
u
M
1
0
q
3
= u
2
0 q
4
= u
2
u
M
2
0
(26)
The proposed cost index that interprets the dis-
cussed need is:
J =
1
2
Z
t
f
t
0
A
1
E
2
+ A
2
I
2
+ B
1
u
2
1
+ B
2
u
2
2
dt (27)
where A
i
and B
j
, i, j = 1, 2 are the weights of the
states, E and I, and of the controls u
1
and u
2
respec-
tively. The optimal control problem can be solved by
using classical techniques of calculus of variation; the
Hamiltonian function is defined as:
H =
1
2
A
1
E
2
+ A
2
I
2
+ B
1
u
2
1
+ B
2
u
2
2
+λ
T
(t)F(X,U) (28)
where λ(t) =
λ
1
(t) λ
2
(t) λ
3
(t)
T
is the costate.
The necessary conditions of optimality are given by:
˙
λ
i
=
H
X
i
, λ
i
(t
f
) = 0 i = 1, 2, 3 (29)
0 =
H
u
j
+
4
k=1
q
k
u
j
η
k
, j = 1, 2 (30)
η
j
q
j
= 0, η
j
0, j = 1, . . . , 4 (31)
Note that in (29) the independence of q
j
, j =
1, ..., 4 from the state X has been used. By taking into
account the conditions (31) the possible 2
4
cases are
reduced to 4. By solving conditions (30) and taking
into account the constraints (26) (5), along with (29),
the optimal controls u
i
, i = 1, 2 are determined:
u
1
= max{min{
(λ
2
λ
4
)σ
1
E
(S + E)B
1
, u
M
1
}, 0} (32)
ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics
458
u
2
= max{min{
(λ
3
λ
4
)σ
2
I
B
2
, u
M
2
}, 0} (33)
Note that the optimal controls require the solution
of the costate equations (29) in addition to the state
equation (7) with the initial conditions (6).
3 NUMERICAL RESULTS AND
DISCUSSION
In this Section, a numerical analysis is performed to
study the proposed computer virus spread model and
the optimal control strategies. For the choices of the
model parameters the value proposed in (Xu and Ren,
2016) are assumed:
b = 5; β = 0.15; µ = 0.3;
ε = 0.01; γ = 0.4 (34)
The initial conditions are set equal to
X
0
=
100 5 0 0
T
(35)
thus assuming a ”population” of 105 devices, among
which 5 are in the E condition of being infected and
infectious, without knowing it. The control period is
set equal to 10 unit of time.
Figure 1: Case 1, β = 0.15: evolution of the number of sus-
ceptible computers (continuous line: free evolution; dotted
line: optimal control is applied).
Two cases are considered, depending on the value
of the probability p. As Case 1 it is chosen p = 0.8,
thus assuming that a computer, after being infected,
with high probability enters in the I class, meaning
that the infection is immediately known, and therefore
it can not infect other computers. With the chosen
values, R < 1 is obtained, thus meaning that only the
equilibrium point X
e1
exists:
X
e1
=
17 0 0 0
T
(36)
By evaluating the corresponding Jacobian (18) and its
eigenvalues, the stability of (36) can be verified. In
Figure 2: Case 1, β = 0.15: evolution of the number of
exposed computers (continuous line: free evolution; dotted
line: optimal control is applied).
Figure 3: Case 1, β = 0.15: evolution of the number of
infected computers (continuous line: free evolution; dotted
line: optimal control is applied).
Figure 4: Case 1, β = 0.15: evolution of the optimal control
strategies (continuous line: u
1
; dotted line: u
2
).
this case the parameters are such that for β = 0.15 it
results β
b
= 0.1826 and then β < β
b
, thus confirming
that the system is below the bifurcation condition with
the chosen parameter values. By changing only the
value of β into 0.25, one gets β > β
b
and a completely
different situation it is obtained: R = 1.37 > 1 and
therefore also the second equilibrium point X
e2
exists:
X
e2
=
12 0 4 0
T
(37)
By considering the Jacobian (17) calculated in the two
Optimal Control to Limit the Propagation Effect of a Virus Outbreak on a Network
459
Figure 5: Case 1, β = 0.25: evolution of the number of sus-
ceptible computers (continuous line: free evolution; dotted
line: optimal control is applied).
Figure 6: Case 1, β = 0.25: evolution of the number of
exposed computers (continuous line: free evolution; dotted
line: optimal control is applied).
Figure 7: Case 1, β = 0.25: evolution of the number of
infected computers (continuous line: free evolution; dotted
line: optimal control is applied).
points X
e1
, (36), and X
e2
, (37), and evaluating the cor-
responding eigenvalues, one has that for X
e1
the Ja-
cobian has one eigenvalue with positive real part, and
therefore it is unstable, while for X
e2
all the eigen-
values have negative real parts, thus guaranteeing its
local asymptotic stability. These results confirm what
has been previously stated in Section 2. The Case
2 regards the choice p = 0.2; this means that after
the infection most of the computers become laten; of
Figure 8: Case 1, β = 0.25: evolution of the optimal control
strategies (continuous line: u
1
; dotted line: u
2
).
Figure 9: Case 2, β = 0.15: evolution of the number of sus-
ceptible computers (continuous line: free evolution; dotted
line: optimal control is applied).
Figure 10: Case 2, β = 0.15: evolution of the number of
exposed computers (continuous line: free evolution; dotted
line: optimal control is applied).
course this is the most dangerous situation. In this
case β
b
= 0.051; by choosing β = 0.15, R = 2.53 > 1
is obtained. Therefore the second equilibrium point
X
e2
exists and assumes the value
X
e2
=
6 4 7 0
T
(38)
Since R > 1 the equilibrium point X
e1
, given again
by (36), is unstable, as it can be verified by calculat-
ing the eigenvalues of (18), whereas the same analy-
sis applied to the second equilibrium point X
e2
in (38)
ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics
460
Figure 11: Case 2, β = 0.15: evolution of the number of
infected computers (continuous line: free evolution; dotted
line: optimal control is applied).
Figure 12: Case 2, β = 0.15: evolution of the optimal con-
trol strategies (continuous line: u
1
; dotted line: u
2
).
guarantees its stability.
If the parameter β is reduced to β = 0.015, that is
to a value lower than β
b
, R = 0.25 < 1 is obtained.
Therefore, a unique stable equilibrium point, X
e1
, is
expected, as it can be directly verified by considering,
as usual, the eigenvalues of (18).
The optimal control actions are determined by
minimizing (27); the values A
1
= 10, A
2
= 0.1, B
1
= 1
and B
2
= 10 are chosen for the weights, whereas for
the boundaries (5) of u
1
and u
2
, u
M
1
= u
M
2
= 1 are as-
sumed. This means that the main aim is to minimize
the number of exposed computers, being the most
dangerous; therefore the limitations of the resources
in the cost index is less restrictive when referring to
the u
1
, rather than u
2
. In Figs. 1–3 the evolutions
of the state in Case 1 with the choice β = 0.15 are
shown; they are the effects of the optimal control de-
picted in Fig. 4, that proposes the maximum allowed
value 1 for u
1
up to about time 5.5, whereas the sec-
ond control u
2
must be at its maximum value up to
time 3. The effects of these controls are evident in the
almost halved number of infected, Fig. 3. As seen, the
same Case 1 analysed with β = 0.25 yields a differ-
ent kind of solution with two equilibrium points; the
evolutions of the state along with the optimal controls
are shown in Figs. 5–8. Due to the more dangerous
situation, the number of exposed and of the infected
devices in free evolution is significantly higher than
in the previous case, Fig. 7, reaching for the infected
computers almost the peak of 65 computers versus 55
of the previous case; also the number of exposed com-
puters reaches higher value (16 versus 13) compen-
sated with the controls, Fig. 6. As far as the control
is concerned, in this case a larger effort, up to time
6, is requested for the control u
1
that at the end of
the control time 10 has not yet reached its minimum
allowed value; similar considerations can be applied
to the optimal control u
2
, that after keeping the maxi-
mum value up to time 3, it decreases without reaching
the minimum value at time 10.
In Case 2, after the infection, a larger number
of computers becomes exposed, being able to infect
other devices; it can be expected a stronger control
action to face this dangerous situation, see Fig. 12 in
case of β = 0.15: for the u
1
action, the maximum ef-
fort is required for almost all the time period, whereas
the maximum value for u
2
is required almost up to
time 4. The results are shown in Figs. 9–11. As ex-
pected, the number of infected computers increases
sensibly in absence of any action but the optimal con-
trol is able to face the spread, Fig. 11. The same Case
2 is analysed assuming β = 0.015; this means that
the virus is not so dangerous as in the previous case;
coherently, the optimal control does not require the
maximum value for both u
1
and u
2
, see Fig. 16: the
best action for u
1
is the maximum value up to time 7,
whereas for the control u
2
it is not required the maxi-
mum value. This is reasonable, since the epidemic is
not spreading and therefore there is not a great need
of anti–virus action. The reduced capability of the
virus to spread can be evidenced also noting that the
evolution of the susceptible devices does not vary sig-
nificantly without and with the control, Fig. 13, being
the epidemic situation not so dangerous.
Figure 13: Case 2, β = 0.015: evolution of the number of
susceptible computers (continuous line: free evolution; dot-
ted line: optimal control is applied).
Optimal Control to Limit the Propagation Effect of a Virus Outbreak on a Network
461
Figure 14: Case 2, β = 0.015: evolution of the number of
exposed computers (continuous line: free evolution; dotted
line: optimal control is applied).
Figure 15: Case 2, β = 0.015: evolution of the number of
infected computers (continuous line: free evolution; dotted
line: optimal control is applied).
Figure 16: Case 2, β = 0.015: evolution of the optimal con-
trol strategies (continuous line: u
1
; dotted line: u
2
).
4 CONCLUSIONS
The computer virus spread represents a problem in
a globalized world more and more connected, being
able to cause economic and social damages. By us-
ing the same modeling of epidemic disease spread, it
is possible to describe the dynamics of the computer
virus and to act in the most efficient way, in order to
avoid as much as possible the obvious remedy, the
isolation of the infected computers. A simple dynam-
ical model to describe the computer virus spread is
proposed and optimal strategies are developed. The
first results appear satisfactory and the effort will be
devoted to apply the discussed model and the control
scheme to a real scenario.
REFERENCES
Driessche, P. V. D. (2002). Reproduction numbers and
sub-threshold endemic equilibria for compartmental
models of disease transmission. Mathematical bio-
sciences, 18.
Driessche, P. V. D. (2017). Reproduction numbers of infec-
tious disease models. Infectious disease modelling, 2.
Fatima, U., Ali, M., Ahmed, N., and Rafiq, M. (2018). Nu-
merical modeling of susceptible latent breaking-out
quarantine computer virus epidemic dynamics. He-
liyon, 4.
Hu, Z., Wang, H., Liao, F., and Ma, W. (2015). Stability
analysis of a computer virus model in latent period.
Chaos, Solitons and Fractals, 75(-):20–28.
Karunanithi, D., Alheyasat, O., Divya, T., and Kavitha, G.
(2018). Attacks on artificial intelligence applications
through adversarial image. International Journal of
Pure and Applied Mathematics, 118(20):4491–4495.
Peng, M., He, X., Huang, J., and Dong, T. (2013). Mod-
eling computer virus and its dynamics. Mathematical
Problems in Engineering, 2013.
Xu, Y. and Ren, J. (2016). Propagation effect of a virus
outbreak on a network with limited anti-virus ability.
PLOS ONE, October(27):1–15.
Zhu, Q. and Yang, X. (2012). Modeling and analysis of
the spread of computer virus. Commun. Nonlinear Sci
Numer Simulation, 17.
ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics
462